2.4 discussion and conclusions 34
2.4 discussion and conclusions
Obtaining reproducible and in-focus images is of utmost importance for further image analysis.
WSSs have built-in procedures to calibrate illumination settings to make the ac- quisition conditions reproducible. Recently, some
WSSmanufacturers provided scanners with new "continuous" focusing methods, that enable to focus each
FOVwhile maintain- ing scanning speed. Although these new focussing methods provide relatively good results with
H&Estained slides, they seem less efficient with other stains, including
IHC. In contrast,
WSSs of the previous generation, such as the one in use at DIAPath, use fast- focusing procedures that estimate a reasonable focus plane from few focusing points.
Although this method work well most of the time, we observed that approximately 25%
of the slides from the DIAPath routine scanning had to be rescanned. To prevent invalid quantification results, a quality control step to assess image sharpness was introduced.
During the elaboration of this quality control step, we found that focusing problems were visible only at magnifications of 10X and above, forcing the operator to assess
VSs sharpness at high magnifications. Due to the size and the amount of the
VSimages to assess, this quality control step was considered as a tedious task, strongly reducing the scanning throughput. To accelerate this task and alleviate scanner operators’ work- load, we developed an automated tool using a supervised classification method. This method proved to be efficient and reduced the strain of the operators during the sharp- ness assessment, while augmenting the scanning throughput of the complete analysis workflow (see Figure 11) in comparison to the manual method.
However, the throughput could be increased even more by interfacing our method with the scanner driver, as carried out in [40], to automatically define new focusing points and scanning areas (see Figure 1C of the paper). The computation of the features used for the classification could also be accelerated using a parallel implementation.
However, our first effort to parallelize the feature computations on graphics processing unit (
GPU) were not conclusive. Indeed, the time needed to transfer the large amount of data to the
GPUoutweighed any benefit of the parallelization.
Finally, even when continuous focussing methods become widespread, a quality con-
trol step will remain necessary before publishing the
VSs. Our tool should thus remain
relevant in this new technological context.
2.4 discussion and conclusions 35
Figure 10: Example of a pre-scanning step result with automatic tissue detection, focal plane division (maximum 1.5 mm length horizontally or vertically) and focusing point posi- tioning (9 per focal plane).
Acquisition device
calibration Image acquistion
Image quality control (sharpness)
Manual or automated ROI definition
Staining characterization
Image registration Staining colocalization Morphological features
extraction
Staining segmentation Statistical analysis on
patient cohort